Download Action, time and the basal ganglia - Philosophical Transactions of

Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
Action, time and the basal ganglia
Henry H. Yin1,2
1
Department of Psychology and Neuroscience, and 2Department of Neurobiology, Center for Cognitive
Neuroscience, Duke University, PO Box 91050, Durham, NC 27708, USA
rstb.royalsocietypublishing.org
Review
Cite this article: Yin HH. 2014 Action, time
and the basal ganglia. Phil. Trans. R. Soc. B
369: 20120473.
http://dx.doi.org/10.1098/rstb.2012.0473
One contribution of 14 to a Theme Issue
‘Timing in neurobiological processes: from
genes to behaviour’.
Subject Areas:
behaviour, cognition, neuroscience, physiology,
computational biology, theoretical biology
Keywords:
dopamine, basal ganglia, bradykinesia, action,
substantia nigra, striatum
Author for correspondence:
Henry H. Yin
e-mail: [email protected]
The ability to control the speed of movement is compromised in neurological
disorders involving the basal ganglia, a set of subcortical cerebral nuclei that
receive prominent dopaminergic projections from the midbrain. For
example, bradykinesia, slowness of movement, is a major symptom of Parkinson’s disease, whereas rapid tics are observed in patients with Tourette
syndrome. Recent experimental work has also implicated dopamine (DA)
and the basal ganglia in action timing. Here, I advance the hypothesis
that the basal ganglia control the rate of change in kinaesthetic perceptual
variables. In particular, the sensorimotor cortico-basal ganglia network
implements a feedback circuit for the control of movement velocity. By
modulating activity in this network, DA can change the gain of velocity
reference signals. The lack of DA thus reduces the output of the velocity control system which specifies the rate of change in body configurations,
slowing the transition from one body configuration to another.
1. Introduction
Although there is no specialized sensory organ for time, our perception of time
nevertheless depends on the rate of change in different sensory modalities. In
the absence of sensory input, e.g. under anaesthesia, the sense of time is
often lost or impaired. That timing depends on the rate of change in perceptual
variables may seem trivially true, but the implications of this proposition are
seldom acknowledged. Here, I shall explain some of these implications in the
kinaesthetic domain by focusing on the timing of actions.
When we move, our body changes its posture, the configurations of different body parts. Not only can we maintain specific body configurations, we can
also control how quickly they change. Although this aspect of behaviour is
often neglected, it becomes more conspicuous in neurological disorders,
which commonly feature deficits in action timing. In Parkinson’s disease, for
instance, movement is often slowed dramatically, a condition known as bradykinesia. By contrast, Tourette syndrome is associated with symptoms like
involuntary and rapid tics. The hypokinetic or hyperkinetic symptoms therefore
suggest that a fundamental action timing mechanism is impaired in these disorders.
The deficits of action timing in both Parkinson’s disease and Tourette syndrome implicate dopaminergic projections to the basal ganglia, a set of
subcortical nuclei in the cerebrum (figure 1a) critical for the learning and
initiation of actions [1 –5]. In Parkinson’s disease, the dopamine (DA) neurons
die, resulting in reduced DA in the basal ganglia. Effective treatments for
bradykinesia include DA replacement therapy with L-DOPA, a DA precursor.
By contrast, Tourette syndrome is thought to result from excessive dopaminergic signalling in the dorsal striatum, and treatments include antagonists of
DA receptors like haloperidol [6]. Slow movements are therefore associated with reduced dopaminergic signalling; fast movements with excessive
dopaminergic signalling.
2. Deficits in timing following dopamine depletion
Clinical symptoms can only provide clues to the underlying mechanisms. To
assess the role of DA in action timing, we experimentally manipulated dopaminergic signalling in mice and used the duration of lever presses as a
measure of action timing. We trained mice on an operant task, in which they
& 2014 The Author(s) Published by the Royal Society. All rights reserved.
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
(a)
cerebral cortex
pallidum
diencephalon
nomifensine in
normal mice
(b)
(c) L-DOPA/carbidopa rescue
300
mean press
duration (ms)
mean press
duration (ms)
300
200
100
0
10
depleted saline
depleted 10/10 mg kg–1
depleted 50/50 mg kg–1
200
100
0
e
–1
n
ali
s
excitatory
inhibitory
modulatory
tectum/brain stem
g
m
–1
kg
20
g
kg
m
(d )
Figure 1. DA and action timing. (a) Schematic of the basic cortico-basal
ganglia circuit. Cortical projection neurons, the pyramidal cells in layer 5, are
glutamatergic and excitatory. Striatal and pallidal projection neurons are by
contrast GABAergic and usually inhibitory. The modulatory dopaminergic projections target most components of the cortico-basal ganglia network, but by
far the largest proportion target the striatum. (b) Intraperitoneal injection of
nomifensine, a blocker of DA transporter, dose-dependently reduced duration
of lever presses in normal mice (n ¼ 5, p , 0.05). All mice were food
deprived and maintained at approximately 85% of normal body weight.
They were trained on a continuous reinforcement schedule for 4 days (each
lever press earned a food pellet). They were then trained on a FI-60 schedule
for at least 5 days before testing (1 h daily sessions). (c) DA depletion in the
sensorimotor striatum increased the duration of lever presses, but this increase in
press duration can be rescued with intraperitoneal injections of L-DOPA/
carbidopa (n ¼ 5, p , 0.0001). (d ) A coronal section of the mouse brain
with tyrosine hydroxylase staining showing selective depletion of DA in the lateral sensorimotor striatum after local 6-OHDA injections (20 mg ml21, 0.5 ml per
side to target dopaminergic terminals bilaterally in the striatum). Note that the
lesion is selective, showing depletion limited to the sensorimotor striatum.
Stereotaxic coordinates relative to bregma in mm: þ0.5, ML + 3.0 and 23
in dorsolateral striatum).
must press a lever for food pellets. First, we used nomifensine
to inhibit DA uptake and increase DA level. We found that
nomifensine dose-dependently reduced press duration
(figure 1b). We then selectively induced DA depletion in
sensorimotor or lateral striatum using 6-hydroxydopamine
(6-OHDA), a toxin that kills DA neurons. DA depletion in
the sensorimotor striatum reduced rate of lever pressing
and increased press duration (figure 1c), in agreement with
previous work [7] as well as clinical observations on
3. Movement as changes in body configuration
A body configuration is the geometrical relationship of body
parts to each another. It is more commonly called a posture,
though there is a persistent misunderstanding of posture as
essentially a static phenomenon that does not require the
active engagement of the brain. But as shown by centuries
of clinical observations, the control of posture or body configuration is an essential function of the nervous system,
from the spinal cord to the brain. It is an active control process that requires the production of continuous neural
outputs coordinated to counter environmental disturbances.
When such control is impaired, a variety of neurological
symptoms result [8]. In fact, impairments in posture and
body configuration are among the most common and
obvious symptoms of neurological disorders.
Movement, then, is but the shifting from one body configuration to another [9]. Movement speed, or the rate of
change in body configurations, can be controlled by the nervous system. This capacity I shall call ‘movement velocity
control’. Some clarifications are needed to avoid any
misunderstanding:
(i) Movement velocity control refers only to the rate of
change in body configurations, though velocity control also applies to perceptions in other modalities,
e.g. visual or auditory configurations.
(ii) A distinction should also be made between movement
velocity and actual velocity of the body in space, as in
locomotion. Riding a stationary bike, for example,
requires high movement velocity, even though the
bike does not move.
(iii) Although movement velocity can be directly controlled, it is not necessarily controlled at all times. It
can also be used in order to achieve the reference conditions of higher systems, in which case the velocity
simply varies as required by the higher level (see
§5 for discussion of how this is possible).
(iv) The perceptual signal representing velocity cannot
be equated with the rate of change in muscle length,
as detected by first-order sensors such as muscle
spindles. With very simple body geometry, stretch
receptors and movement velocity detectors may be
equivalent. But in most cases, the rate of change in
body configuration requires a higher order representation of a collection of lower order kinaesthetic
inputs. However it is implemented, for our purposes
it is sufficient to assume only that movement velocity
can be detected and controlled by the nervous system.
Phil. Trans. R. Soc. B 369: 20120473
DA cell
groups
2
rstb.royalsocietypublishing.org
striatum
bradykinesia in Parkinson’s patients. This deficit can also
be rescued by DA replacement treatment with L-DOPA
(figure 1c). These results suggest that dopaminergic innervation, particularly that of the sensorimotor striatum, is
critical for the timing of actions.
Action duration reflects the rate of change in kinaesthetic
variables. Following DA depletion, it appears that movement
velocity, i.e. rate of transition from one body configuration to
another, cannot exceed a certain level. These observations raise
the question of how DA and the basal ganglia can implement
action timing. Before attempting to answer this question,
however, it helps to define action timing more explicitly.
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
4. Closed-loop control
The nervous system is hierarchically organized, implementing multiple levels of control [9–11]. Different negative
feedback control systems can be arranged hierarchically, the
output of a higher level system serving as a reference signal
that specifies the input that the lower control system must
obtain. This basic principle distinguishes the hierarchical
relationship proposed here from other hierarchical models
in neuroscience and psychology which are all open loop
and thus incapable of controlling anything [12,13].
Since each control system is defined by its controlled variable, different levels of the hierarchy control different
variables. To control the rate of change in body configuration,
it is necessary first to control body configuration, which in
turn requires the control of muscle length and tension.
6. Bradykinesia
Deficits in action timing can be understood as a result of
altered signalling in specific parts of the hierarchy just
described. In bradykinesia, for example, it is as if the configuration reference signal were low-pass filtered, changing
very slowly. The most extreme example is akinesia, in
which the configuration is simply fixed. It is important to
note that the lower body configuration control system is
still functioning, still producing resistance to any disturbance
to body configuration, regardless of how slowly the velocity
reference signal changes, if at all. Such resistance is observed,
for example, in the so-called ‘lead pipe rigidity’, found in
Parkinson’s patients.
Phil. Trans. R. Soc. B 369: 20120473
5. Hierarchical control
3
rstb.royalsocietypublishing.org
Control systems, as defined here, bring the value of a
particular input variable closer to a desired value, despite
environmental disturbances. For example, room temperature can be controlled by a thermostat, just as body
temperature is controlled by the nervous system, despite fluctuations in the outside temperature. For this to be possible, it
is necessary to sense the value of the input variable to be controlled, and to have an internal representation of the desired
value of the variable. The sensor can send a signal representing the value of the controlled variable to a comparator,
which computes the difference (error signal) between the
input and a reference signal representing the desired value
of the controlled variable. The error signal is transformed
into the appropriate output.
A velocity control system, then, can detect how rapidly
its input signal is changing, compare the sensed speed
with a reference speed, and adjust the output until the
sensed speed matches the reference signal. For this to be
possible, the output must affect sensed velocity through a
feedback function which expresses the input signal as a
mathematical function of the output signal. The loop is
thus closed in the environment. Negative feedback reduces
the error signal so that the desired value of the controlled
variable can be reached. This organization, it should
be noted, is the only one found to be effective for
controlling the value of a given variable despite environmental disturbances. In practice, no engineer would rely
on any other method, e.g. feed-forward mechanisms, for
effective control.
One example of velocity control is cruise control in cars,
where a negative feedback mechanism is used. Yet how
such control is achieved by the brain is different. In biological
control systems, the reference signal is intrinsic to the
organism, not accessible to external users as it is in manmade control systems like the thermostat. The organism is
self-adaptive, able to tune its own parameters. The userspecified ‘input’ is eliminated in natural control systems.
Moreover, for any moderately complex body geometry,
first-order sensors cannot adequately sense movement velocity. A simple negative feedback control system is
therefore not sufficient. Rather than sending its output to
the muscles directly, the velocity control system can only
exert indirect effects on the muscles. Instead, a hierarchy of
control systems is required.
In other words, it is necessary to use additional lower
level control systems to exert effects on the environment.
As the controlled variable must be sensed, each level of
the hierarchy is defined by its sensory input (figure 2). Movement velocity control is largely independent of the distal
senses such as vision and hearing. The primary sensory
modality required is kinaesthetic. The relevant sensory
inputs are ultimately derived from signals from proximal sensors in the body. Such signals, like all sensory signals, are
transformed by successive levels of the hierarchy into representations of higher order variables such as movement
velocity. It is this higher order input variable that is fed
into the velocity control system and compared with the
velocity reference signal.
Still higher levels can also use velocity control for their
purposes by sending their reference signals that are proportional to their own error signals. For example, when
tracking a moving target, the proximity between the hand
and the target is an error signal in a higher level that controls
the relationship between the two. Such a relationship control
system requires visual inputs. Closing the distance reduces
the relationship error, which can be transformed into a reference signal for the velocity control system. But this can be
achieved in different ways, such as reaching with the arm
or walking towards the moving object, which involve transitions in different body configurations. Either way, it is
possible to control the proximity between hand and object
by varying the reference signal for velocity.
Proportional to the velocity error signal is the output of
the velocity control system, which must send reference signals to a lower control system for body configuration or
posture. A body configuration can be conceived as a collection of joint angles. At each joint, antagonistic muscles can
pull in opposite directions, their relative lengths determining
the joint angle. The simplest configuration is that of one joint,
but for normal movements multiple joints are involved.
The movement velocity control system is involved in
transitioning from one body configuration to another.
It determines how much time it takes to change from one
configuration to another and how long a particular configuration is maintained. It can reach the desired velocity
by varying the reference signal to the body configuration
system. The transition from one configuration to another is
normally continuous and smooth as the reference signal
changes. I propose that a velocity error signal is transformed
into the rate of change in the configuration reference signal.
This transformation is similar to the mathematical operation
of leaky integration [14].
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
comparator
velocity input
velocity error
input
output (with leaky integration)
configuration reference
comparator
input
kinaesthetic signals
configuration error
output
muscle length control
Figure 2. Hierarchical implementation of velocity control. The hierarchical
relationship between velocity control and body configuration control situated
immediately below in the hierarchy. The error signal from the velocity control
system is converted into a reference signal for the body configuration control system. With integration in the output function of the velocity control
system, the velocity error signal is proportional to the rate of change in
the body configuration reference. The higher the velocity reference, the
faster the change in body configuration.
In principle, bradykinesia can be produced by increased
sensitivity to velocity input, reduced velocity reference
signal or altered velocity output. Exaggerated velocity perception can be produced by an increase in input gain, so
that velocity is perceived as faster than it actually is. That
is, given a particular velocity, the velocity sensor sends a
larger than normal signal to the comparator. With the same
reference signal, the error signal in the velocity system will
be reduced, and less output (lower rate of change in configuration reference) is produced. Should this be the case, small
variations in velocity could produce large effects on behavioural output, which would show high-frequency noise,
high frequencies having larger effects on sensed velocity.
If, on the other hand, bradykinesia is produced
by reduced gain in the function that transforms the error
signals from higher levels into the velocity reference signal,
then a smaller velocity reference signal is produced. Thus,
velocity control is functioning normally, though the reference
signal it receives is always low (which is also the case
when one deliberately tries to move slowly). The maximum
value the reference signal attains during a movement is
reduced. With an integrator in its output function, the
output of the velocity control system—the reference signal
for configuration/position control will change slowly.
Finally, the problem may be found in the transformation
of a normal velocity error into a reference signal for configuration control, in the generation of the time integral.
For example, if the integrator becomes too leaky, the
configuration reference signal can also change more slowly.
The above possibilities are not mutually exclusive. From
clinical observations alone, it is difficult to rule out any one
of them [15]. Selective manipulation of the relevant neural circuits will be necessary to determine the specific mechanisms
underlying bradykinesia. The outlining of these possibilities
above will serve to illustrate the reasoning used in analysing
properties of hierarchical control systems.
4
DA in the basal ganglia appears to play a critical role in the
control of movement velocity. As many comprehensive
reviews on the organization of the basal ganglia are available
[16 –18], only the most relevant features are outlined here.
The basal ganglia are a group of subcortical nuclei with
highly conserved circuitry [19– 21]. As shown in figure 1,
unlike the cerebral cortex, which contains excitatory glutamatergic projection neurons, the basal ganglia contain
g-aminobutyric acid (GABA) projection neurons [18]. The
main input nucleus is the striatum, a large and heterogeneous
region including the dorsal striatum (caudate and putamen in
primates) and ventral striatum (nucleus accumbens). The lateral septum and parts of the central amygdala are also
classified as striatal regions on account of their GABAergic
projection neurons. The main output nucleus is the pallidum,
which includes entopeduncular nucleus in rodents or globus
pallidus internus in primates, and substantia nigra pars reticulata [18]. These ‘pallidal’ structures send projections to
the tectum, brainstem and thalamus.
The striatum is the major target of dopaminergic projections, with the highest density of DA receptors in the brain.
It also receives extensive and topographically organized glutamatergic projections from the cortex and thalamus [22– 24].
The medium spiny striatal projection neurons are the primary
targets of the glutamatergic and dopaminergic projections.
Rather than depolarizing or hyperpolarizing the target
neurons, the activation of G protein-coupled DA receptors
modulates the excitatory effect of glutamate, depending on
the subtype of DA receptors expressed on the target neuron
[25,26].
An important feature of the basal ganglia is the existence
of two neuronal populations in the striatum giving rise to the
so-called direct and indirect pathways, one projecting to the
substantia nigra pars reticulata (SNr) and the internal segment of the globus pallidus (GPi) and the other projecting
to the external segment of the globus pallidus (GPe), which
in turn projects to the SNr. These two populations differ in
the type of DA receptor expressed: the direct or striatonigral
pathway expresses D1-type DA receptors whereas the indirect or striatopallidal pathway expresses D2-type receptors
[16,27]. These two pathways can thus exert opposite effects
on the SNr neurons [28]. The direct pathway has a net inhibitory effect, while the indirect pathway has a net excitatory
effect on the SNr outputs [28]. Because the output of the
SNr is inhibitory, the activation of the direct pathway is
expected to disinhibit downstream structures, whereas that
of the indirect pathway produces the opposite effect [28,29].
8. Neural activity related to action duration
To study the role of the basal ganglia in action timing, we
developed an operant duration differentiation task in mice
[30 –34]. On each trial, to earn a piece of food reward, the
mouse must produce an action of a minimum duration
[32,33]. The lever is transiently retracted after release, and a
reward is delivered if the press duration exceeded the criterion duration. No cue tells the animal whether the action
duration is long enough. Only after lever retraction is the outcome revealed, and the presentation of the reward is the only
way that the animal learns about the efficacy of its lever
Phil. Trans. R. Soc. B 369: 20120473
configuration input
7. Dopamine and the basal ganglia
rstb.royalsocietypublishing.org
velocity reference
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
9. Targets of the basal ganglia
The outputs of the nigral GABAergic neurons can reach the
tectum (superior colliculus), the ventral lateral and ventral
anterior thalamus, the pedunculopontine nucleus and the
reticular formation [16,17,41,42]. If basal ganglia output represents a configuration reference signal, then among the
major targets of these projections there should be control systems for body configurations. The rate of change in this
reference signal will then determine how quickly one body
configuration changes to another.
It is possible that different targets in the diencephalon
and brainstem mediate different aspects of body configuration control. For example, locomotion requires relatively
stereotyped transitions like alternation of limb flexors and
extensors, whereas changes in posture require mainly transitions in the proximal musculature. The SNr sends strong
projections to the mesopontine tegmentum, a brainstem
region critical for initiation of locomotion and regulation of
posture [43]. The lateral SNr projects to the pedunculopontine
nucleus, whereas the medial SNr projects to the mesencephalic
locomotor region. They can be the targets for the pair of
opponent nigral output reference signals discussed above.
Whether these structures implement the body configuration
control systems remains to be determined.
The nuclei in the mesopontine tegmentum in turn project
to the reticulospinal system, the major motor pathway in all
vertebrates [19,44].
Axons from the reticular formation in the brainstem reach
motor neurons innervating different muscles all over the
5
Phil. Trans. R. Soc. B 369: 20120473
The same reasoning can be applied to the antagonistic
body configuration systems, which are hierarchically higher
than the spinal reciprocal inhibition circuits.
A body configuration control system can simultaneously send signals to multiple muscles, adjusting their lengths
via the activation of g-motor neurons. The higher order perceptual signal could be a weighted sum of signals from
multiple muscle length detectors as well as other sensors
for joint angles. Thus, more global opponent signals can be
found at a higher level, controlling the lengths of groups of
muscles, instead of the local opponent signals sent to two
muscles doing work in opposite directions with respect to a
single joint. For example, to pull the door open, a set of
muscles (e.g. biceps) does the pulling while a different set
of muscles (leg extensors) pushes against the ground.
The value of a controlled variable like muscle tone can
vary from zero to some maximum, and a viable operating
point would be in the middle of this range, to permit variations above and below the average value. If the tone is too
low, then there is no room for the signal to decrease. If it is
too high, there is no room for any increase.
Likewise, the high tonic firing rate of the GABAergic
output neurons in the SNr may also allow neural activity in
this basal ganglia output nucleus to increase and decrease,
generating a pair of opponent signals [39,40].
It is hypothesized that, for any body configuration,
opponent reference signals may be needed. In fact, at lower
levels, the reference signals should come in pairs. But at
higher levels, the velocity reference signal is singular, and it is
only in the transformation from velocity reference to position/
configuration that a pair of reference signals is generated.
rstb.royalsocietypublishing.org
pressing (figure 3a). In other words, the reward delivery is
contingent upon press duration, rather than press rate, as in
traditional operant conditioning [32,35,36].
Press duration can be used to tag the neural activity generating the action itself [33,35]. The body configuration
reference signal reaches and stays at a steady-state value as
long as the lever is held down. With the release of the
lever, the output of the velocity control system changes the
reference signal for the configuration control system again.
To obtain reward, the instrumental contingency requires the
velocity reference to be very low for the duration of the action.
Using temporal differentiation, we studied the neural
activity of GABAergic output neurons in the SNr and of the
nearby dopaminergic neurons in the pars compacta [35].
We found two major neural activity patterns in relation to
the lever press action in the recorded neurons: in one type
of neurons (‘action-on’) there was a sustained increase in
firing rate during the action, and in the other population
(‘action-off’) there was a sustained pause in firing rate
during the press (figure 3). These two types are found in
both GABAergic and dopaminergic neurons, though it is
not clear how these types relate to previously identified
heterogeneous neuronal populations in the nigra [37].
As long as the lever is held down, the reference signal for
the velocity control system should be close to zero, while the
reference signal for body configuration remains unchanged.
Once a body configuration is reached, it is maintained for
the duration of the action. With an integrator in the output
function, the velocity control system can send a relatively
constant reference signal for body configuration. The rate of
change is low when a posture is maintained. The observed
action-related neural activity can reflect either the velocity
or body configuration reference signal.
But we found two signals that are similar in amplitude
but opposite in polarity. Although the pause of the ‘actionoff’ neurons can represent zero velocity during the action,
the increased output of the ‘action-on’ neurons does not
appear to represent the velocity reference signal. Nor does
it seem plausible to send a high velocity reference signal
and a zero velocity reference signal at the same time.
Another possibility is that the opponent outputs reflect
descending body configuration reference signals for antagonistic lower control systems. But why should there be a pair of
signals, rather than a single reference signal? The most wellknown example of opponent signals in the nervous system is
reciprocal innervation in the spinal motor neurons. The
output of a given a-motor neuron innervates a particular
muscle, but at the same time a branch of this output excites
the Ia interneuron, which inhibits the a-motor neuron
innervating the antagonist muscle. Thus, a pair of signals is
sent to the effectors, one with a net excitatory effect on a
muscle and the other with a net inhibitory effect on the
antagonist muscle. This arrangement is needed because
muscles can only pull. Antagonism between muscles is
defined with respect to a particular joint: e.g. the biceps
will pull in one direction, reducing the joint angle; whereas
the triceps will pull in the opposite direction, increasing the
joint angle. To reduce the angle of the elbow joint, it is not
only important for the biceps to contract but also for the
triceps to relax. At rest, there is a balance of activity in agonist
and antagonist motor neurons. Simultaneous contraction of
these muscles increases net angular stiffness or impedance
at a joint [38].
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
(a)
6
release
press
rstb.royalsocietypublishing.org
criterion
duration
rewarded trial
unrewarded trial
no reward
Phil. Trans. R. Soc. B 369: 20120473
(b)
reward
(c)
average waveforms
DA
GABA
spikes s−1
(d )
80
40
0
–1
0
1
time from press start (s)
20
10
0
2 –1
0
1
time from press start (s)
2
Figure 3. Opponent outputs from the basal ganglia. (a) Illustration of the duration differentiation task [33,35]. Mice must press a lever and hold it down for a
minimum amount of time in order to earn a food pellet. Only after the lever is released is the trial outcome revealed (reward or no reward). (b) Illustration of the
electrode implant into the substantia nigra. (c) The positioning of the electrode array allows simultaneous recording of activity from dopaminergic and GABAergic
neurons, which can be distinguished on the basis of their action potential waveforms. (d ) Raster plots showing typical ‘action-on’ and ‘action-off’ GABAergic neurons. Each row is a single trial from a temporal differentiation session. Yellow markers indicate ‘lever start’; red markers indicate ‘lever end unrewarded’; green
markers indicate ‘lever end rewarded’. The trials are sorted according to the duration of the lever press, starting with trials with the shortest action durations
on top.
body. Interestingly, opponent activity is also observed in the
reticulospinal tract [45,46]. Instead of agonist and antagonist
muscles defined with respect to single joints, groups of
muscles probably send kinaesthetic signals to the reticulospinal system [44].
Although the mesopontine tegmentum and the reticulospinal system are discussed here, the other targets of the
basal ganglia output (including the tectum and ventral thalamus) may also contain body configuration control systems
[17,41,47].
10. Lower levels for motor control
Below the level of the body configuration control, there are at
least two additional levels (figure 4). But, partly because of
technical limitations, not much is known about the functional
organization of the spinal cord where these lower levels are
located. The reticulospinal pathway can send reference
signals representing desired muscle lengths for specific
muscles, probably by activating spinal g-motor neurons and
interneurons. The g-motor neurons can in turn send reference
signals for muscle length, which are compared with the input
from muscle spindles. Finally, the a-motor neurons serve as
comparators that receive force reference signals from multiple
sources. Their output is roughly proportional to the degree of
muscle contraction. Whether there is co-activation of g- and
a-motor neurons, as is often claimed [48], remains unclear,
because in a closed-loop system signals cannot be considered
seriatim as in verbal description.
The muscle is the output function of the force control
system, the lowest level in the motor hierarchy that closes
the loop in the environment. Muscle tension is sensed by
the Golgi tendon organs, which send Ib afferents representing
the negative feedback for the force control system. In this connection, it is worth noting that increased stiffness in skeletal
muscles is also observed in Parkinson’s patients, as a result
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
input function
comparator
output function
DA
transition/velocity
striatonigral MSN
body configuration
PPN/MLR
g-motor neuron
muscle length
muscle spindle
muscle tension
a-motor neuron
reticular formation
intrafusal muscle
la afferent
extrafusal muscle
external environment
Figure 4. Possible neural implementation of the control hierarchy. A highly simplified illustration of the hierarchy for movement velocity control. Note that DA
neurons also receive strong projections from the body configuration control system. The tectum and parts of the ventral thalamus should also belong to the body
configuration control system. They might specialize in configurations of specific body parts, e.g. head and neck. PPN, pedunculopontine nucleus; MLR, mesencephalic
locomotor region; MSN, medium spiny neuron.
of increased muscle tone in both agonist and antagonist
muscles. Such rigidity can also reduce movement velocity,
because mechanical impedance is the ratio between velocity
and force needed to achieve that velocity, which can be
increased by simultaneous activation of agonist and antagonist muscles [38]. This can be achieved by a descending
reference signal to both types of motor neurons, a mechanism
that overrides reciprocal inhibition. Parkinsonian rigidity,
however, is sometimes relieved when the patient is suspended in water, suggesting that it is manifested when
posture must be defended against gravity [15]. Thus, rigidity
may well be a compensatory reaction to the loss of normal
control functions for body configurations.
11. The role of dopamine in the basal ganglia
Based on the above outline of basal ganglia anatomy and
review of experimental data, I propose that the sensorimotor
cortico-basal ganglia network is responsible for controlling
transitions between body configurations and plays a critical
role in movement speed (figure 4). The role of the basal
ganglia in movement speed has long been recognized [49–52],
yet the mechanisms for hierarchical control have never been
incorporated in a model to explain action timing.
The perceptual input signal representing the rate of
change in body configuration is the velocity feedback. It
may be carried by the projections to the sensorimotor striatum from sensorimotor cortical regions, the intralaminar
thalamus and the globus pallidus [53]. On the other hand,
reference signals for movement velocity may be sent via the
cortical and thalamic projections to the sensorimotor or lateral striatum. In vivo recording in mice has also shown high
correlation between movement velocity on a rotarod and
the firing of the sensorimotor striatal and motor cortical neurons [54]. Results from DA depletion experiments discussed in
§2 suggest that dopaminergic innervation of the sensorimotor
striatum, in particular, is critical for movement velocity control (figure 1). Selective depletion in this region is sufficient
to increase the duration of normal actions by reducing gain
in the velocity control system.
For higher levels in a hierarchy, the output function transforms a given error signal into a reference signal for the level
below, i.e. an error in proximity between one’s body and the
target in tracking movements is transformed into a reference
signal for the desired velocity for the hand. As a neuromodulator, DA cannot produce significant synaptic currents
directly to produce firing. But it can change the gain of
other signals, in this case the reference signal for the rate of
change, which is sent via the glutamatergic projections to
the striatum [55]. Reduced DA may therefore result in a
reduced velocity reference signal. This proposed mechanism
is in agreement with the known function of DA in modulating striatal synaptic transmission [25,56,57]. As a result of
DA depletion, the velocity reference signal is reduced. The
control system receiving such a signal produces a low
output entering the integrator, resulting in a slow rate of
change in the body configuration reference signal. When
the velocity reference signal is zero, then the configuration
reference signal will no longer change, resulting in a lack of
voluntary movement. This condition, akinesia, is seen in
severe cases of Parkinsonism.
Phil. Trans. R. Soc. B 369: 20120473
joint angle
SNr
rstb.royalsocietypublishing.org
relationship reference
7
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
12. Conclusion and implications
8
Phil. Trans. R. Soc. B 369: 20120473
Observable behaviour consists of transitions between body
configurations. It is hypothesized here that movement velocity—rate of change of body configurations—is controlled
by the sensorimotor cortico-basal ganglia circuit. It does not
follow that velocity control is the sole function of the basal
ganglia. Nor does the present hypothesis exclude other
roles of DA in behaviour. But it does suggest that there is a
common computational function performed by the basal
ganglia networks.
A striking feature of the cortio-basal ganglia networks
is the uniformity in the basic circuitry. All cortical areas
have a similar structure, as do all striatal areas and all pallidal
areas. Different networks are simply variations of a common
motif: e.g. excitatory outputs of the cerebral cortex and thalamus, inhibitory outputs of the basal ganglia, and modulation
of synaptic transmission and neuronal excitability by transmitters such as DA. This motif has been conserved in the
evolution of vertebrates [19]. By comparison, variations in
the expression of receptors and cytoarchitectonic features
are minor.
On the other hand, there is also much functional heterogeneity in the cortico-basal ganglia networks. Lesions to
different regions produce different effects on behaviour
[58]. The question is how we can reconcile this functional
heterogeneity, the ‘localization of function’ known since
antiquity, and the proposed common computational function
of the basal ganglia.
What is normally called function is related to the behavioural consequences of lesion or stimulation of a brain
region, determined by the overall connectivity of the region.
For example, visual deficits are a consequence of lesions to
the primary visual cortex. But a distinction must be made
between ‘function’ in this sense and computation, which is
what happens to signals entering a neural circuit. The
neural signal is firing rate, an analogue signal, in spite of
the misleading all-or-none digital property of the generation
of individual action potentials [59]. Computations using these
signals can be described with a set of mathematical operations often used in analogue computing, such as addition,
subtraction and integration.
The difference, say, between the limbic cortico-basal
ganglia network and the sensorimotor network lies not in
the type of computation implemented by the neural circuits,
but in the content of signals, i.e. what they represent. As control is always the control of input, the identity of the
controlled variable depends solely on its input to the control
system. The sensorimotor striatum, for example, receives
higher order kinaesthetic and somesthetic signals from the
primary sensorimotor cortices, whereas a limbic striatal
region such as the nucleus accumbens shell receives a different set of more poorly defined inputs from areas like the
basolateral amygdala, a very different cortical region [60].
Exactly what the controlled variables are for different
brain circuits remains unclear. So far only speculations are
possible based on the known anatomy and the behavioural
deficits following lesions. Given the tremendous heterogeneity in the connectivity of different cortical, striatal and
pallidal regions, elucidating the content of the variables will
require extensive experimental work designed to test the
specific controlled variables.
Here, a hypothesis is advanced regarding the sensorimotor cortico-basal ganglia network, based on the DA depletion
and in vivo recording data. But the basic circuit for transforming a velocity error signal into a configuration or position
reference signal is still the same. A velocity reference signal
enters the basal ganglia circuit which then generates
opponent configuration reference signals that are sent to the
level below. One possible neural implementation is suggested
by the anatomy: the excitatory input to the striatum activates
both the striatonigral and striatopallidal pathways [61], and
the intrinsic organization of these two pathways transforms
the uniform excitation into a pair of signals, one increasing
and the other decreasing from a common mode signal. And
finally, a time integral can be produced in the basal ganglia
transforming a given magnitude of velocity error into a rate
of change in the reference signal for body configuration control.
If negative feedback control of velocity is accomplished
by the sensorimotor cortico-basal ganglia circuit, then the
other networks (e.g. limbic and associative) must be responsible for controlling different input variables [21,62]. In this
light, different cortico-basal ganglia networks can be
viewed as distinct higher level control systems for the control
of the rate of change in different higher order perceptual variables, of which body configuration in only one example. The
number of possible controlled variables is virtually unlimited, as different sensory variables can be combined to form
higher order variables representing abstract categories (e.g.
reward), especially in the cerebral cortex.
In conclusion, control of the rate of change or transitions
is perhaps the biological basis for our sense of time, even
though time itself may not be a directly controlled variable.
Much research has implicated the basal ganglia and DA in
timing, but the underlying mechanisms remain unclear [63].
While the timing mechanism can be relatively independent
of actual movements, it must still depend on the higher
level of control of transitions between perceptual configurations which implicate the basal ganglia and modulation
by DA (figure 4).
I have attempted to incorporate the known anatomy of
the nervous system in a model of the control hierarchy, linking velocity control with higher levels such as relationship
control, and lower levels such as body configuration control,
muscle length control and muscle force control. Clearly,
much of the model remains speculative, as many relevant
facts are still unknown. But even at this early stage some
predictions can be made. First, glutamatergic input to the
sensorimotor striatum and its dopaminergic modulation
play a critical role in movement velocity. Manipulations of
the glutamatergic signal alone, or of the firing of striatonigral
projection neurons, should systematically affect movement velocity. Second, the basal ganglia outputs from the
substantia nigra pars reticulata should contain reference
signals for body configuration. As it is a general property
of successful control systems that inputs match reference
signals, activity in basal ganglia output nuclei should be
rstb.royalsocietypublishing.org
Increased dopaminergic signalling in the sensorimotor
cortico-basal ganglia network produces the opposite effect.
Even a small relationship error signal, which is not normally
sufficient to produce any movement velocity reference, is
now greatly amplified. Consequently, ‘uncontrollable’ rapid
movements may be produced. Such symptoms appear to be
common in Tourette syndrome.
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
Acknowledgements. The author thanks Mark Rossi for comments and
suggestions on the manuscript.
Funding statement. The author is supported by NIH AA021074.
References
1.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18. Swanson LW. 2000 Cerebral hemisphere regulation
of motivated behavior. Brain Res. 886, 113 –164.
(doi:10.1016/S0006-8993(00)02905-X)
19. Grillner S, Wallen P, Saitoh K, Kozlov A, Robertson B.
2008 Neural bases of goal-directed locomotion in
vertebrates—an overview. Brain Res Rev. 57,
2 –12. (doi:10.1016/j.brainresrev.2007.06.027)
20. Luo M, Ding L, Perkel DJ. 2001 An avian basal
ganglia pathway essential for vocal learning forms
a closed topographic loop. J. Neurosci. 21,
6836 –6845.
21. Yin HH, Knowlton BJ. 2006 The role of the basal
ganglia in habit formation. Nat. Rev. Neurosci. 7,
464 –476. (doi:10.1038/nrn1919)
22. Berendse HW, Groenewegen HJ. 1990 Organization
of the thalamostriatal projections in the rat, with
special emphasis on the ventral striatum. J. Comp.
Neurol. 299, 187–228. (doi:10.1002/cne.902990206)
23. Ding J, Peterson JD, Surmeier DJ. 2008
Corticostriatal and thalamostriatal synapses have
distinctive properties. J. Neurosci. 28, 6483–6492.
(doi:10.1523/JNEUROSCI.0435-08.2008)
24. Ragsdale Jr CW, Graybiel AM. 1991 Compartmental
organization of the thalamostriatal connection in
the cat. J. Comp. Neurol. 311, 134–167. (doi:10.
1002/cne.903110110)
25. Kreitzer AC. 2009 Physiology and pharmacology of
striatal neurons. Annu. Rev. Neurosci. 32, 127– 147.
(doi:10.1146/annurev.neuro.051508.135422)
26. Surmeier DJ, Ding J, Day M, Wang Z, Shen W. 2007
D1 and D2 dopamine-receptor modulation of
striatal glutamatergic signaling in striatal medium
spiny neurons. Trends Neurosci. 30, 228–235.
(doi:10.1016/j.tins.2007.03.008)
27. Gerfen CR, Surmeier DJ. 2011 Modulation of striatal
projection systems by dopamine. Annu. Rev.
Neurosci. 34, 441 –466. (doi:10.1146/annurevneuro-061010-113641)
28. Hikosaka O, Takikawa Y, Kawagoe R. 2000 Role of
the basal ganglia in the control of purposive saccadic
eye movements. Physiol. Rev. 80, 953–978.
29. Kravitz AV, Freeze BS, Parker PR, Kay K, Thwin MT,
Deisseroth K, Kreitzer AC. 2010 Regulation of
parkinsonian motor behaviours by optogenetic
control of basal ganglia circuitry. Nature 466,
622 –626. (doi:10.1038/nature09159)
30. Platt JR, Kuch DO, Bitgood SC. 1973 Rats’ leverpress durations as psychophysical judgements of
time. J. Exp. Anal. Behav. 19, 239 –250. (doi:10.
1901/jeab.1973.19-239)
31. Skinner BF. 1938 The behavior of organisms.
New York, NY: Appleton-Century-Crofts.
32. Yin HH. 2009 The role of the murine motor cortex in
action duration and order. Front Integr. Neurosci. 3,
23. (doi:10.3389/neuro.07.023.2009)
33. Yu C, Gupta J, Yin HH. 2010 The role of mediodorsal
thalamus in temporal differentiation of rewardguided actions. Front. Integr. Neurosci. 4, 14.
(doi:10.3389/fnint.2010.00014)
34. Zeiler MD. 1985 Pure timing in temporal
differentiation. J. Exp. Anal. Behav. 43, 183–193.
(doi:10.1901/jeab.1985.43-183)
35. Fan D, Rossi MA, Yin HH. 2012 Mechanisms of
action selection and timing in substantia nigra
neurons. J. Neurosci. 32, 5534– 5548. (doi:10.1523/
JNEUROSCI.5924-11.2012)
36. Rossi MA, Sukharnikova T, Hayrapetyan VY, Yang L,
Yin HH. 2013 Operant self-stimulation of dopamine
neurons in the substantia nigra. PLoS ONE 8,
e65799. (doi:10.1371/journal.pone.0065799)
37. Henny P, Brown MT, Northrop A, Faunes M, Ungless
MA, Magill PJ, Bolam JP. 2012 Structural correlates
of heterogeneous in vivo activity of midbrain
dopaminergic neurons. Nat. Neurosci. 15, 613 –619.
(doi:10.1038/nn.3048)
38. Hogan N. 1984 Adaptive control of mechanical
impedance by coactivation of antagonist muscles.
Autom. Control IEEE Trans. 29, 681–690. (doi:10.
1109/TAC.1984.1103644)
39. Rossi MA, Fan D, Barter JW, Yin HH. 2013
Bidirectional modulation of substantia nigra activity
by motivational state. PLoS ONE 8, e71598. (doi:10.
1371/journal.pone.0071598)
40. Zhou FM, Lee CR. 2011 Intrinsic and integrative
properties of substantia nigra pars reticulata
neurons. Neuroscience 198, 69 –94. (doi:10.1016/j.
neuroscience.2011.07.061)
41. Garcia-Rill E. 1986 The basal ganglia and the
locomotor regions. Brain Res. 396, 47 –63. (doi:10.
1016/0165-0173(86)90009-3)
42. Niijima K, Yoshida M. 1982 Electrophysiological
evidence for branching nigral projections to pontine
reticular formation, superior colliculus and
thalamus. Brain Res. 239, 279–282. (doi:10.1016/
0006-8993(82)90852-6)
43. Takakusaki K. 2008 Forebrain control of locomotor
behaviors. Brain Res. Rev. 57, 192 –198. (doi:10.
1016/j.brainresrev.2007.06.024)
44. Peterson BW. 1979 Reticulospinal projections to
spinal motor nuclei. Annu. Rev. Physiol. 41, 127–
140. (doi:10.1146/annurev.ph.41.030179.001015)
45. Deliagina TG, Beloozerova IN, Zelenin PV, Orlovsky
GN. 2008 Spinal and supraspinal postural networks.
Brain Res. Rev. 57, 212 –221. (doi:10.1016/j.
brainresrev.2007.06.017)
46. Orlovsky GN, Deliagina TG, Wallen P. 1992
Vestibular control of swimming in
lamprey. I. Responses of reticulospinal neurons to
roll and pitch. Exp. Brain Res. 90, 479– 488.
(doi:10.1007/BF00230930)
Phil. Trans. R. Soc. B 369: 20120473
2.
Buse J, Schoenefeld K, Munchau A, Roessner V.
2012 Neuromodulation in Tourette syndrome:
dopamine and beyond. Neurosci. Biobehav. Rev. 37,
1069–1084. (doi:10.1016/j.neubiorev.2012.10.004)
Graybiel AM. 1998 The basal ganglia and chunking
of action repertoires. Neurobiol. Learn. Mem. 70,
119–136. (doi:10.1006/nlme.1998.3843)
Hallett M, Khoshbin S. 1980 A physiological
mechanism of bradykinesia. Brain 103, 301 –314.
(doi:10.1093/brain/103.2.301)
Yin HH. 2010 The sensorimotor striatum is necessary
for serial order learning. J. Neurosci. 30, 14 719 –
14 723. (doi:10.1523/JNEUROSCI.3989-10.2010)
Yin HH, Ostlund SB, Knowlton BJ, Balleine BW. 2005
The role of the dorsomedial striatum in
instrumental conditioning. Eur. J. Neurosci. 22,
513–523. (doi:10.1111/j.1460-9568.2005.04218.x)
Worbe Y, Malherbe C, Hartmann A, Pelegrini-Issac
M, Messe A, Vidailhet M, Lehericy S, Benali H. 2012
Functional immaturity of cortico-basal ganglia
networks in Gilles de la Tourette syndrome. Brain
135, 1937 –1946. (doi:10.1093/brain/aws056)
Cousins MS, Sokolowski JD, Salamone JD. 1993
Different effects of nucleus accumbens and
ventrolateral striatal dopamine depletions on
instrumental response selection in the rat.
Pharmacol. Biochem. Behav. 46, 943–951. (doi:10.
1016/0091-3057(93)90226-J)
Martin JP. 1967 Basal ganglia and posture.
Philadelphia, PA: Lippincott.
Bernstein N. 1996 Dexterity and its development.
Mahwah, NJ: Lawrence Erlbaum Associates.
Cools A. 1985 Brain and behavior: hierarchy of feedback
systems and control of input. In Perspectives in ethology,
pp. 109–168. Berlin, Germany: Springer.
Powers WT. 1973 Behavior: control of perception.
New Canaan, CT: Benchmark.
Botvinick MM, Niv Y, Barto AC. 2009 Hierarchically
organized behavior and its neural foundations: a
reinforcement learning perspective. Cognition 113,
262–280. (doi:10.1016/j.cognition.2008.08.011)
Fuster JM. 1995 Memory in the cerebral cortex.
Cambridge, MA: MIT press.
Robinson DA. 1989 Integrating with neurons. Annu.
Rev. Neurosci. 12, 33 –45. (doi:10.1146/annurev.ne.
12.030189.000341)
Sacks O. 1991 Awakenings. London, UK: Pan
Macmillan.
Gerfen CR, Wilson CJ. 1996 The basal ganglia.
In Handbook of chemical neuroanatomy, vol. 12 (eds
LW Swanson, A Bjorklund, T Hokfelt), pp. 371–468.
Amsterdam, The Netherlands: Elsevier.
Hikosaka O. 2007 GABAergic output of the
basal ganglia. Progr. Brain Res. 160, 209–226.
(doi:10.1016/S0079-6123(06)60012-5)
9
rstb.royalsocietypublishing.org
correlated with actual body configuration. It is hoped that
predictions such as these may be useful in guiding future
experimental endeavours.
Downloaded from http://rstb.royalsocietypublishing.org/ on May 11, 2017
59.
60.
61.
62.
63.
accumbens: the integrative functions of corticobasal ganglia networks. Eur. J. Neurosci. 28, 1437–
1448. (doi:10.1111/j.1460-9568.2008.06422.x)
Von Neumann J. 2012 The computer and the brain.
New Haven, CT: Yale University Press.
Swanson LW. 2003 The amygdala and its place in
the cerebral hemisphere. Ann. NY Acad. Sci. 985,
174–184. (doi:10.1111/j.1749-6632.2003.tb07081.x)
Cui G, Jun SB, Jin X, Pham MD, Vogel SS,
Lovinger DM, Costa RM. 2013 Concurrent
activation of striatal direct and indirect pathways
during action initiation. Nature 494, 238 –242.
(doi:10.1038/nature11846)
Redgrave P, Prescott TJ, Gurney K. 1999 The basal
ganglia: a vertebrate solution to the selection
problem? Neuroscience 89, 1009 –1023. (doi:10.
1016/S0306-4522(98)00319-4)
Buhusi CV, Meck WH. 2005 What makes us tick?
Functional and neural mechanisms of interval
timing. Nat. Rev. Neurosci. 6, 755–765. (doi:10.
1038/nrn1764)
10
Phil. Trans. R. Soc. B 369: 20120473
53. Kita H, Kita T. 2001 Number, origins, and chemical
types of rat pallidostriatal projection neurons. J. Comp.
Neurol. 437, 438–448. (doi:10.1002/cne.1294)
54. Costa RM, Cohen D, Nicolelis MA. 2004 Differential
corticostriatal plasticity during fast and slow motor
skill learning in mice. Curr. Biol. 14, 1124–1134.
(doi:10.1016/j.cub.2004.06.053)
55. Hernandez G, Breton Y-A, Conover K, Shizgal P.
2010 At what stage of neural processing does
cocaine act to boost pursuit of rewards? PLoS ONE 5,
e15081. (doi:10.1371/journal.pone.0015081)
56. Servan-Schreiber D, Printz H, Cohen JD. 1990 A
network model of catecholamine effects: gain,
signal-to-noise ratio, and behavior. Science 249,
892 –895. (doi:10.1126/science.2392679)
57. Thurley K, Senn W, Luscher HR. 2008 Dopamine
increases the gain of the input-output response of
rat prefrontal pyramidal neurons. J. Neurophysiol.
99, 2985–2997. (doi:10.1152/jn.01098.2007)
58. Yin HH, Ostlund SB, Balleine BW. 2008 Rewardguided learning beyond dopamine in the nucleus
rstb.royalsocietypublishing.org
47. Hess WR. 1957 The functional organization of the
diencephalon. New York, NY: Grune & Stratton.
48. Granit R. 1970 The basis of motor control:
integrating the activity of muscles, alpha and
gamma motoneurons and their leading control
systems. London, UK: Academic Press.
49. Anderson M, Horak F. 1985 Influence of the globus
pallidus on arm movements in monkeys. III. Timing
of movement-related information. J. Neurophysiol.
54, 433–448.
50. Bullock D, Grossberg S. 1989 VITE and FLETE: Neural
modules for trajectory formation and postural control. In
Volitional action (ed. W Hershberger), pp. 253–298.
Amsterdam, The Netherlands: Elsevier/North Holland.
51. Horak FB, Anderson ME. 1984 Influence of globus
pallidus on arm movements in monkeys. I. Effects of
kainic acid-induced lesions. J. Neurophysiol. 52,
290–304.
52. Horak FB, Anderson ME. 1984 Influence of globus
pallidus on arm movements in monkeys. II. Effects of
stimulation. J. Neurophysiol. 52, 305–322.